Login flow over-fragments into 12+ internal steps for a single user-step

[hardikshah197]

Summary

When a *_test.md file specifies a single user-step like "navigate to URL X, fill email, fill password, click Login", the kane-cli agent fragments it into 12+ internal steps. Each internal step burns ~10-30s and 3-6 credits.

Example trace

For a one-line "go to dev site and log in" objective, the agent ran:

1. navigate → dev.lambdatestinternal.com
2. cert warning Advanced
3. cert warning Proceed
4. wait (analyze if page loaded)
5. wait (analyze again)
6. wait (analyze if login form visible)
7. analyze visual: is login form present?
8. click email field
9. fill email
10. click password field
11. fill password
12. click Login button
13. wait
14. analyze URL fragment for access_token=
15. assert

For a 7-step test plan I authored, the agent expanded it to ~80+ internal steps and 200+ credits.

Observed behavior

• 7-step test_md → ~80 internal steps → 263 credits → 21 min wall time.
• A trivial single-step "log in and confirm Insights loaded" → ~15 internal steps → ~50 credits.

Suggested fixes

• Provide a primitive login-like action that batches cert-bypass + form fill + submit + wait-for-redirect into one step.
• Recognize repeated form-fill patterns (click→fill→click→fill→submit) and collapse them.
• Cert-warning interstitial click-through trips the agent on every fresh run — surface a flag like --auto-bypass-cert-warning to handle it without per-test steps.

Environment

• kane-cli 0.2.9
• macOS 14, Chrome 148